evo-memory

Pass

Audited by Gen Agent Trust Hub on Mar 15, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The skill operates entirely within the local workspace environment, using standard file manipulation tools (read_file, write_file, edit_file) to maintain research history. No external network requests or unauthorized data access patterns were found.\n- [PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection as it ingests and processes data from external research artifacts.\n
  • Ingestion points: Research proposals, tournament direction summaries (/direction-summary.md), and experiment trajectory logs (/experiments/*/trajectory.md).\n
  • Boundary markers: The skill employs structured reasoning steps and specific prompts for extraction, but does not explicitly define character-based delimiters (e.g., XML tags or triple quotes) for untrusted content in its templates.\n
  • Capability inventory: Capabilities are restricted to reading/writing workspace files and agent reasoning (think_tool).\n
  • Sanitization: No specific sanitization or filtering of input data is implemented beyond the LLM's intrinsic reasoning processing.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 15, 2026, 10:52 AM